Download chapter i - Trace: Tennessee Research and Creative Exchange

Document related concepts

Introduction to gauge theory wikipedia , lookup

Fundamental interaction wikipedia , lookup

Lepton wikipedia , lookup

Electrical resistivity and conductivity wikipedia , lookup

Condensed matter physics wikipedia , lookup

Standard Model wikipedia , lookup

State of matter wikipedia , lookup

Strangeness production wikipedia , lookup

History of subatomic physics wikipedia , lookup

Elementary particle wikipedia , lookup

Plasma (physics) wikipedia , lookup

Transcript
To the Graduate Council:
I am submitting here with a thesis written by Densu Aktas entitled “Enhancement of PROBEPIC
Code and Application to Langmuir Probes.” I have examined the final electronic copy of this
thesis for form and content and recommend that it be accepted in partial fulfillment of the
requirements for the degree of Master of Science, with a major in Mechanical Engineering.
Trevor M. Moeller, Major Professor
We have read this thesis
and recommend its acceptance:
Roy Schulz
Gary Flandro
Accepted for the Council:
Carolyn R. Hodges
Vice Provost and Dean of the Graduate School
(Original signatures are on file with official student records.)
Enhancement of PROBEPIC Code and
Application to Langmuir Probes
A Thesis Presented for
The Master of Science
Degree
The University of Tennessee, Knoxville
Densu Aktas
May 2010
Copyright © 2010 by Densu Aktas
All rights reserved.
ii
DEDICATIONS
This thesis is dedicated to my Mother,
Sara Alexis
Whom I love very much
iii
“Happy indeed is the scientist who not only has the pleasures which I have enumerated, but
who also wins the recognition of fellow scientists and of the mankind which ultimately benefits
from his endeavors.“
-Irving Langmuir, inventor of the Langmuir
Probe, also coined the term "plasma".
iv
ACKNOWLEDGEMENTS
This thesis would not have been possible without the support of my friends and my
grandma. I also want to thank the members of the First Baptist Church in Nashville, Tennessee,
who provided important financial and spiritual support, especially Mr. Morris Early “grandpa.”
I would like to thank my major advisor, Dr. Trevor M. Moeller, for his assistance and
guidance, and my committee for their patience and support.
I would also like to thank my mentor and friend, Mr. Paul Gloyer. Without his help, I
never would have been able to make it through all of the challenges that I faced in the last two
years. His guidance and council could never be repaid. He served as a wonderful example and
showed me how to be a professional engineer.
Tanri hepsini korusun ve kutsasin.
v
ABSTRACT
The Langmuir probe is used as a diagnostic devise to determine the properties of
plasma, which is a state of matter that is partially or wholly ionized gas containing free
charged particles; free moving electrons and ions. This probe is used to determine plasma
characteristics such as temperature and density by carefully analyzing the probe voltagecurrent (V-I) characteristics. However, real world effects, such as finite length geometry, cause
the measurements from these probes to deviate from the theoretical predictions for ideal
probes.
To understand these discrepancies and predict the performance of real probes,
computer simulations using the Particle-in-Cell (PIC) and Monte Carlo techniques have been
used. These simulations provide insight to help understand experiments that can provide
detailed information that is difficult to measure, and help design new experiments and probes.
The current effort expands upon the thesis work performed by Thomas Markusic that
resulted in the creation of a 2D/3V PIC code, PROBEPIC, in 1996 at UTSI, to simulate a
Langmuir probe to analyze the plasma behavior. In the current effort, Markusic’s code is
updated from C to C++ and also converted into FORTRAN. In addition to resolving coding issues,
vi
this effort identified several opportunities to enhance the effectiveness of the tool. By adding a
ray tracing model and modifying the boundary condition models, the accuracy of the tool was
improved.
Chapter I provides an introduction to the effort, including a discussion of plasma
characteristics and the Langmuir probe theory. This chapter also contains brief information
about Markusic’s simulation, code upgrades and the code conversion process. Chapter II is
focused on the description of the PIC code and the improvements and updates made in
PROBEPIC. The issues encountered during the creation of the new version of the PROBEPIC
code are also addressed along with their solutions. Chapter III presents the results that show
the conversion was successful and that the new results are in agreement with both previous
data and theoretical predictions. Finally, Chapter IV contains conclusions and some
recommendations for future work.
vii
TABLE OF CONTENTS
CHAPTER I ....................................................................................................................................... 1
Introduction and General Information .......................................................................................... 1
1.1
Langmuir Probe Description .......................................................................................... 4
1.1.1 Langmuir Probe .......................................................................................................... 5
1.1.2 Thomas Markusic’s Simulation ................................................................................ 14
1.2
PIC Monte Carlo Description ....................................................................................... 17
1.3
Approach ...................................................................................................................... 21
1.3.1 Transformed Probe PIC to FORTRAN and the Changes Made ................................. 21
CHAPTER II ................................................................................................................................... 24
Description of POBEPIC ................................................................................................................ 24
2.1
Details on Probe PIC ..................................................................................................... 25
2.2
Improvements/Updates ................................................................................................. 27
2.3
Comparison of New/Old Langmuir Simulation ........................................................... 39
CHAPTER III .................................................................................................................................. 40
Results ........................................................................................................................................... 40
CHAPTER IV .................................................................................................................................. 45
Conclusions and Recommendations ............................................................................................. 45
4.1
Conclusions ................................................................................................................... 45
4.2
Recommendations ......................................................................................................... 46
BIBLIOGRAPHY ............................................................................................................................. 47
APPENDICES ................................................................................................................................. 49
Vita ................................................................................................................................................ 51
viii
LIST OF TABLES
Table 1: Quiescent plasma simulation conditions in PROBEPIC [2] ............................................... 17
Table 2: PROBEPIC sub-program description [2] ................................................................................... 23
Table 3: A sample of editable parameter.txt ............................................................................................. 36
ix
LIST OF FIGURES
Figure 1: A simple Langmuir probe set up [6] ............................................................................................ 7
Figure 2: Schematic of a typical Langmuir probe ..................................................................................... 8
Figure 3: Typical Voltage-Current (V-I) characteristic for a Langmuir probe ............................. 9
Figure 4: The view of the objects in a typical PIC code [2] ................................................................. 15
Figure 5: A typical cycle, a onetime step, in a particle simulation program (PIC) .................... 20
Figure 6: PROBEPIC computational domain layout created [2] ........................................................ 26
Figure 7: Field boundary conditions and probe geometry in PROBEPIC....................................... 27
Figure 8: The computational grid in PROBEPIC (before correction) .............................................. 28
Figure 9: The computational grid after the correction of grid.c in PROBEPIC ........................... 29
Figure 10: Particle ray tracing shown in animation window............................................................ 32
Figure 11: The rectangular domain in the original PROBEPIC ......................................................... 33
Figure 12: Cylindrical domain boundary with particle injection over 90 degrees .................. 34
Figure 13: The locations where the particles injected and hit the probe...................................... 38
Figure 14: Voltage-Current (V-I) characteristics of original and updated PROBEPIC code . 41
Figure 15: Probe grid with 50 percent lower density ........................................................................... 44
x
CHAPTER I
Introduction and General Information
Plasma can be defined as quasi-neutral ionized gas with a roughly equal
number of positively and negatively charged particles. This collection of free-moving
electrons and ions makes electromagnetic effects important for the plasma physical
behavior. Unlike gasses, liquids or solids, plasma does not contain molecules. Instead, it
is composed of ions and accounts for more than 99 percent of the known visible
universe. Even the solar system is permeated with plasma, and the earth is completely
surrounded by plasma trapped within its magnetic field.
The American chemist, Irving Langmuir, was the first scientist to apply the term
“plasma” to describe an ionized gas in the mid-twenties. From his research, Langmuir
invented a new probe device to measure the properties of plasma [1]. Even today, the
Langmuir probe is still an important tool in plasma diagnostics and research.
Furthermore, his basic work with this probe opened the door to the development of
many other methods to measure the composition and thermodynamic properties of
1
plasma, e.g. interferometer (optical and microwave), spectroscopy, spectral intensity
and electrostatic probes [6].
With proper design and careful data analysis techniques, the Langmuir probe
can provide important information concerning plasma parameters. This information
includes the ion density, the plasma electron density, the plasma potential, the plasma
electron temperature, and the plasma electron energy distribution function (EEDF) [8].
Langmuir’s method accomplishes this with the measurement of the electron and ion
currents to a small metal electrode, i.e. Langmuir probe, as different voltages are applied
to the probe. This yields a curve called the probe characteristic of the plasma (described
in section 1.1.1).
Since the masses of electrons and ions are very different, they respond to forces
on different time scales. This, combined with their opposite charges, produces the
varying response seen in the probe characteristic curve. For example, large negative
voltages produce a small magnitude current for the heavy slow positive ions, while
large positive voltages produce a high magnitude current due to the fast electrons.
However in all cases, the number of electrons equals the number of ions (i.e. ne = ni) in
the plasma to satisfy the requirement of overall charge neutrality.
2
While the Langmuir probe theory provides an explanation of the ideal behavior
of a Langmuir probe, the real world effects result in measurements that deviate from the
ideal voltage-current characteristic curve [2]. Since these differences can impact the
data interpretation and understanding of plasma, it is important to have the means to
predict the real world effects prior to testing.
To accomplish this, researchers [2, 3, 7] have applied modern computational
tools to simulate the interaction of a Langmuir probe with plasma. Using particle in cell
(PIC) and Monte Carlo techniques, it is now possible to predict the probe characteristic
curve to some degree of accuracy. A notable example of such a tool is the PROBEPIC
code developed by Thomas Markusic and documented in his Master of Science Thesis
[2]. PROBEPIC extends the work of previous treatments [3] by simulating the behavior
of finite length probes. Markusic used this tool to conduct a parametric study on the
Langmuir probe and interpret available experimental data [2]
The purpose of the current research presented in this thesis is to update and
enhance the PROBEPIC code to improve its accuracy and operation on modern
computers.
Specifically, the code was updated from C language to C++ and then
transformed into FORTRAN to provide two updated versions of the PROBEPIC code. As
3
part of this effort, a variety of minor issues were resolved and simple refinements were
made. Additionally, the current research provided opportunities to refine underlying
assumptions in the original code. These features were addressed by implementing
enhanced models that improve the accuracy of the tool. With the enhanced version of
PROPEPIC, updated voltage-current characteristics for a reference Langmuir probe
were generated, and presented in Chapter III.
1.1
Langmuir Probe Description
As previously mentioned, many fundamental plasma parameters can be
determined by placing a small conducting probe into the plasma and observing the
current to the probe as a function of the difference between the probe and plasma space
potentials (i.e. voltages). This simple electrostatic device is known as a Langmuir probe.
In this section the Langmuir probe theory is explained, and the computational
simulation of the Langmuir probe pursued by Thomas Markusic in PROBEPIC is
described.
4
1.1.1 Langmuir Probe
In plasma, since the electrons have smaller masses, they have higher thermal
speeds than ions at the same temperature. As a result the electrons will migrate to the
probe faster than the ions, causing the probe to collect the electrons at a higher rate.
Eventually, this process causes the floating potential of the probe to become less than
the plasma potential, which is the potential of the plasma with respect to the walls of
the device at a given location [4]. Although plasma potential, Vp, is usually a few volt
positive with respect to the device walls, electron and ion densities are considered to be
essentially equal to each other.
Between the uninterrupted neutral plasma and boundary wall (at a lower
potential than the plasma) there is a region where the electrons are repelled and the
positive ions are attracted. The thickness of this area is called Debye shielding length, λD.
Debye shielding length is a characteristic distance in plasma beyond which the electric
field of a charged particle is shielded by opposite sign charged particles and it is
calculated by the following equation [5].
1⁄
2
ε0 ∗ k ∗ Te
λD = ( 2
)
e ∗ ne
1⁄
2
TeV
= 7.4 ∗ 10 ∗ ( )
ne
3
5
(1.1)
TeV =
k ∗ Te
e
(1.2)
where k is the Boltzmann constant, Te is the electron temperature in Kelvin, TeV is the
electron temperature in electron volts (11,600 K ≡ 1 eV), e is the electron charge and ne
is the electron density in (m-3).
Initially, it was thought that plasma potential could be determined by measuring
the potential on the probe relative to one of the electrodes. However, later on, it was
determined that the plasma potential had to be determined from the floating potential,
Vf, of the probe. Although the plasma potential and the floating potential are completely
different, the real difference between them could not be measured until Irving
Langmuir and Harold Mott Smith provided an experimental method to determine the
plasma potential in 1920s [2, 6]. From this work, they developed the Langmuir Probe to
provide measurements of electron density, temperature and ion density from simple
electric measurements.
A typical Langmuir probe set up uses a ramp voltage that is applied to the probe
positioned in the plasma to generate a probe current, which is typically recorded with
an oscilloscope, to determine the voltage-current (V-I). The data from the oscilloscope is
6
then sent to a computer for further analysis which allows one to determine electron
temperature and density. Figure1 shows the typical experimental set up for Langmuir
probe applications.
Figure 1: A simple Langmuir probe set up [6]
7
In general, the Langmuir probe is nothing more than a sphere, cylinder or a disk
which is made out of a conducting substance immersed into plasma. Figure2 illustrates
the typical Langmuir probe shape. The typical construction of a Langmuir probe
includes two sections, the probe shaft and the probe tip. The probe tip is usually a metal,
such as tungsten, and is normally insulated by mica or quartz. Only a small section of
the probe (i.e. probe tip) is exposed to the plasma and electrically biased with respect to
a reference electrode to collect electron and/or positive ion currents. With this
configuration, the probe is exposed to a large flux of ions and electrons, the magnitude
of which depend upon the density and the temperature of the particles and the voltage
applied to the probe. The charged plasma particles collide with the exposed probe
surface and this allows the Langmuir probe to draw electrical current.
Figure 2: Schematic of a typical Langmuir probe
8
Irving Langmuir’s method consists of obtaining the voltage-current (V-I)
characteristic of the probe when the bias voltage, the voltage between two electrodes, is
swept from negative to the positive potential. Figure3 illustrates the typical currentvoltage characteristic for a Langmuir probe. The probe current-voltage characteristic
shown in Figure3 can be typically divided into several regions as described below [5].
Figure 3: Typical Voltage-Current (V-I) characteristic for a Langmuir probe
9
As seen in Figure3, the floating potential is the point where the current collected
is zero, designated Vf (V=Vf). At this point, the ion and electron currents are equal and
the probe current is zero. The floating potential can be found with the equation below
[5].
1⁄
2
k ∗ Te
mi
Vf = −
∗ ln (
)
e
4 ∗ π ∗ me
(1.3)
where mi is the mass of ion and me is the mass of electron.
If the probe potential is less than the floating potential, V < Vf, then the probe
current is mainly positive ion current. Ion saturation occurs rapidly as V is reduced
below Vf, as shown between points A and B (Figure 3). On the other hand, if the probe
potential is greater than the plasma potential, V > Vp, the result is space charge limited
electron current, indicating electron saturation, as shown between points C and D
(Figure 3). However, if the probe potential is between the floating and the plasma
potential, Vf < V < Vp, then the current is mostly due to electron-diffusion to the probe.
While theory for a probe in ideal plasma indicates that the current in this region would
vary linearly with the probe voltage, in reality the current varies exponentially with the
probe voltage [5].
10
When the probe potential equals the plasma potential, V = Vp, the electrons and
ions diffuse to the probe as if they were unaffected by its presence. The local zero
condition is characterized by the lack of plasma sheath. On either sides of this point, the
difference in voltage introduces electric forces that create a boundary region known as
the plasma sheath, a region where charge neutrality does not exist.
When there is no plasma sheath, the charged particles can migrate to the probe
freely due to their thermal velocities. In this case, it is assumed that the probe current,
Ip, does not interrupt the plasma equilibrium, and the probe diameter, rp, is considered
less than the electron mean free path. Furthermore in this case, the electrons are
considered to be in thermal equilibrium at a temperature Te with a Maxwellian kinetic
energy distribution. In this case the electron current diffusing to the probe can be found
with the following equation [5].
Ip =
1
∗ A ∗ n ∗ e ∗ υ̅
4
(1.4)
where n is the electron density in the immediate vicinity of the probe surface when V <
Vp, A is the surface area of the probe, e is electron charge and 𝜐̅ is average electron
speed. Here, n can be found using the following equation [5].
11
−e ∗ (Vp − V)
(Vp − V)
n = ne ∗ exp (
) = ne ∗ exp (
)
k ∗ Te
k ∗ TeV
(1.5)
where k is Boltzmann constant, Te is electron temperature, Vp is plasma potential (also
called space potential) and V is probe potential. The average electron speed can be
calculated with the equation below [5].
8 ∗ k ∗ Te
8 ∗ e ∗ TeV
υ̅ = √(
) = √(
) = 6.7x105 √TeV ms−1
π∗m
π∗m
(1.6)
Taking the natural logarithm of equation (1.5) and with the help of the equation (1.4),
the following formula can be found [5].
ln Ip = ln Ips −
Vp
V
−
TeV TeV
(1.7)
where Ips is the probe current when V = Vp, and it is calculated using the following
equation [5].
Ips =
1
∗ A ∗ ne ∗ e ∗ υ̅
4
12
(1.8)
Then the slope of the curve lnIp versus probe potential, V, is given by the formula [5].
d ln Ip
1
=
dV
TeV
(1.9)
By differentiating the logarithm of the electron current with respect to the probe
voltage, V, for V < 0, the electron temperature can be obtained. In this case, the slope of
voltage versus lnIp is a straight line in the region between the plasma potential and the
floating potential. It is important to note that the straight line is achieved only if the
distribution is Maxwellian. Again, this slope determines the electron temperature using
the equation (1.9). This equation also shows that the positive ion current is small and
can be neglected in most of the region except very close the Vf [5]. Also using the value
obtained for TeV, the mean electron speed can be obtained from equation (1.6) and the
electron density can be obtained from equation (1.8) [5].
The aim of this sub-section was to give the reader a brief description of the
Langmuir probe and the basic theory to show how it is applied to determine properties
of the plasma. More detailed information can be found in Markusic’s thesis [2].
13
1.1.2 Thomas Markusic’s Simulation
During his graduate study, Thomas Markusic created some Particle-in-Cell
(PIC) simulations to show the behavior of the finite length Langmuir probes [2]. He used
PIC as a computational method to simulate the plasma phenomena. The PIC contains
particles, grids and boundaries. In PIC, many discrete particles are used to simulate the
collective behavior of plasma to come as close to physical reality as possible. A spatial
grid on which the electromagnetic fields are computed is superimposed on the
computational domain. The charges of the particles in the grid cells are weighted to the
grid nodes using Poisson’s equation; the electric field is calculated on the nodes; and the
charge particles are accelerated by the electric field [7]. Figure4 illustrates the objects in
the PIC code.
Markusic’s thesis work was related to probe simulations in quiescent plasmas
and ion beams, and the simulations were shown to be in agreement with both theory
and experimental results [2]. Markusic’s objective was to provide an accurate modeling
of the physical processes using robust algorithms.
14
Figure 4: The view of the objects in a typical PIC code [2]
Based on electric probe theory, approximate solutions were available for two
limiting cases during Markusic’s work. One was for very thin sheaths, and the other
one was for very thick sheaths. According to the thin sheath case, space charge limited
current collection must be assumed. For this case, any particles crossing the sheath are
collected. Moreover, the Debye shielding length is much smaller than the probe radius
(λD<<rp). According to that approximation, the resulting voltage-current characteristic
would have a very sharp knee at point C in Figure3.
15
For the very thick sheath case, orbital motion limited (OML) current collection
must be assumed. OML current is the current collected by the probe when undisturbed
particles are not capable of reaching the probe on the basis of kinetic energy and the
intervening barrier of effective potentials. In this case, the influence of the sheath is
neglected, and the particle orbits are computed using the space charge electric field of
the probe. For this case, the Debye shielding length is much greater than the probe
radius (λD >>rp). With this approximation, the resulting voltage-current characteristic
does not have a very sharp knee (at point C in Figure 3); rather, a very subtle change
from positive to negative curvature occurs.
After examining the two limiting cases, Markusic chose to model the
interactions of a probe with plasma that has properties consistent with the very thick
sheath case, since OML theory was the most relevant to the Langmuir probe theory and
analysis created by I. Langmuir. Table1 shows the plasma simulation conditions used in
PROBEPIC simulation to determine the V-I characteristic and assess its accuracy [2].
16
Table 1: Quiescent plasma simulation conditions in PROBEPIC [2]
Hydrogen and mi/me=100 Conditions
1.2
N [cm-3]
1.0x10^(9)
T [eV]
2.0
Plasma potential [V]
0.0
Probe aspect ratio
45.92
PIC Monte Carlo Description
The Monte Carlo method is a technique that involves using random numbers
and probability to solve problems. Through the Monte Carlo method, many
approximate solutions for a variety of mathematical problems can be performed via
statistical sampling experiments on a computer. The method applies to problems with
absolutely no probabilistic content, as well as to those with inherent probabilistic
structure [9].
Modeling physical problems via Monte Carlo method allows the examination of
very complex systems. Solving equations describing the interactions between two
particles is a very simple process. However, solving the same equations for hundreds or
17
thousands of particles is currently impossible. However, with Monte Carlo methods, a
large system can be sampled in a number of random configurations, and that data can
be used to describe the system as a whole.
For the PROBEPIC version developed in this thesis, standard random number
generators were examined to determine their quality and usability. Based on this
evaluation, the “random_number.c” algorithm used in the original PROBEPIC was
replaced. In the new version of PROBEPIC, “random_number.c” [2] was changed to
“random_number.cpp” for the C++ version and “random_number.f90” for the FORTRAN
version. These improved generators allow one to send an array of any size to be filled
with random numbers of a uniform deviation. This allows one to store many values for
later use rather than calculating them every time they are called in the PROBEPIC
simulation. When all of the random numbers in the array are used, the array will be
sent back to the algorithm random_number.cpp (or random_number.f90) with a new
seed and initialization to generate another array of random numbers [2].
The particle simulation method has been studied for a long time and physicists
working with the simulation of charged particles have developed the particle-in-cell
(PIC) method based on the theory of charge assignment and force interpolation. This
18
important theory has made it possible to obtain smooth distributions of charge density
and current density needed for Maxwell’s equation of electromagnetic fields [7]. PIC
and Monte Carlo plasma simulations have existed separately for several decades, with
PIC applied primarily to hot fully ionized plasmas without collision, and Monte Carlo
method applied to cool weakly ionized gases. However, these capabilities have now been
merged together and can be applied to both plasmas with collisions and to electron flow
in semiconductors [7].
A PIC program contains four main modules which are used to move the particles
through the phase space (velocity and spatial coordinates). These modules are the
particle mover, the charge weighter, the field solver and the force weighter. At each
time step in time, each of these modules is executed sequentially (Figure 5) [7]. In a
characteristic period of the plasma, there may be several thousands of time steps. The
order of these modules is shown in Figure5 [7]. Every particle is moved; the charges and
velocities of the particles are weighted to the grid cells to determine the charge density
and current density on the grid points; the field solver then calculates the
electromagnetic fields on the grid points; these field quantities are used to calculate the
force on each particle; and then the cycle is repeated for the next time step.
19
Charge Weighting
(xi, vi)
(ρj, jj)
Move Particles
Fi
xi
Field Solver
vi
(ρj, jj)
(Ej, Bj)
Force Weighting
(Ej, Bj)
Fi
Figure 5: A typical cycle, a onetime step, in a particle simulation program (PIC) [7]
Monte Carlo simulation is a method for iteratively evaluating a deterministic
model using sets of random numbers as inputs. Therefore, the generation of pseudorandom numbers is an important element of any PIC Monte Carlo simulation. The
quality of the random number plays a very important role in the statistical accuracy of
the simulation. The ability to generate uncorrelated numbers with a uniform deviation
is determined by the quality of a random number generator. Note that the fastest
generators take a lot of space in the computer memory [7, 9].
20
1.3
Approach
The first objective of this thesis work was to translate PROBEPIC, which was
created in C code by T. Markusic in 1996 [2], into FORTRAN to improve its
compatibility with other tools. Next, the effort refined the PROBEPIC code to improve its
accuracy, and upgrades were made to assumptions in the earlier code. To verify the
capabilities of the new PROBEPIC code, it was used to simulate a Langmuir probe
reference case to create a voltage-current (V-I) plot that can be compared with theory
for an ideal probe. An overview of the transformation of PROBEPIC to FORTRAN is
discussed in this section, with a more detailed discussion provided in Chapter II.
1.3.1 Transformed Probe PIC to FORTRAN and the Changes Made
T. Markusic created the PIC code, PROBEPIC, for the simulation of a Langmuir
probe to determine the plasma characteristics as described in his thesis [2]. His
programming philosophy was to develop a concise, intuitively accessible code without
much redundancy. He used the C programming language since it was a popular
language that allowed for longer variable name lengths than other programming
21
languages of the day. C code also allows for a complicated data structure, giving the
programmer control over dynamic memory allocation.
PROBEPIC was comprised of twenty-two individual subprograms, each of which
performs an individual task. He designed these functions to be independent of any
external variables. Table2 shows these independent sub-programs with their functions
listed next to them. To be able to run PROBEPIC with its sub-programs, Markusic was
pushing the computational capabilities of the day requiring. The computations in
Markusic’s thesis were primarily done with the C4 super computer at the AEDC high
performance computer center. However, advances in computer technology now allow
PROBEPIC to be run on a desktop computer in roughly the same amount of time. In the
present effort, the sub-programs given in Table2 were debugged and upgraded to C++
and then they were converted into FORTRAN with documentation so that they can be
run by users familiar with either programming language. During the conversion
process, some issues were encountered; these issues and their solutions are explained in
detail in Chapter II. Appendix1 contains printed listings of these programs in both
computer languages.
22
Table 2: PROBEPIC sub-program description [2]
Sub-program
boundary.c
boundary_beam.c
charge_weight.c
field_solver.c
force_weight.c
graphics.c
grid.c
initialize.c
initialize_beam.c
inject_beam.c
inject_part.c
locate.c
make_LU.c
make_velocity_table.c
maxwell.c
mover.c
output_data.c
parameter.c
probepic.c
probepic.h
random_number.c
reset_grid.c
Function
Handles particle/Boundary interactions for thermal particles
Handles particle/ Boundary interactions for beam particles
The charge weighter
The electric field solver
The force weighter
Graphical diagnostic output
Generates the computational grid
Distributes the initial loading of thermal super particles
throughout the computational domain
Distributes the initial loading of beam super particles
throughout the computational domain
Fluxes beam particles across boundaries into the
computational domain
Fluxes thermal particles across boundaries into the
computational domain
Determines what grid-cell a particle is in
Performs LU decomposition of difference matrix for
subsequent use in field solver
Creates arrays of velocities with for fluxing into the
computational domain
Creates arrays of velocities with for fluxing into the
Maxwellian distribution
The mover
Generates PROBEPIC output file
Determines some necessary parameters used in other subprograms from initial plasma conditions
PROBEPIC main program
Header file for PROBEPIC
Provides uniform random numbers for all PROBEPIC subprograms
Resets charge density, electric field, etc. at grid-points to zero
at beginning of each time-step
23
CHAPTER II
Description of POBEPIC
The charged particles making up plasma move under the influence of the
electromagnetic fields that they generate, together with any external fields that may be
present. Thus, in simulating the evolution of plasmas one seeks to model the mutual
interaction of all particles and fields in a self-consistent manner. The Particle-in-Cell
(PIC) method is a widely used technique for the modeling plasma systems; the method
(detailed in the last chapter) simulates the motion of plasma particles and calculates all
macro-quantities such as number density, current density, etc., from the position and
velocity of these particles [7].
T. Markusic, using the PIC code, desired to simulate plasma movement without
using an experimental set up and predict the results as close as possible to the real
conditions. He coined the name as PROBEPIC in 1996 as the name of the code. With the
present thesis work, his code is updated into C++ and converted into FORTRAN. In this
section, these processes are explained in detail. Additionally, a discussion is provided
24
for the modeling improvements that were incorporated to update assumptions in the
earlier code.
2.1
Details on Probe PIC
Markusic created PROBEPIC, which was a 2D/3V PIC code to simulate the
behavior of a Langmuir probe. 2D/3V means that the electric field has two components
while the particles are free to move in three dimensions [7]. The code contains of
twenty-two independent sub-programs, which are listed in Table2.
The computational domain in PROBEPIC consists of a cylindrical region which
contains a cylindrical Langmuir probe, as illustrated in Figure6. The probe consists of
cylindrical conducting wire partially covered by an alumina insulator [2]. The
computational domain surrounding the probe provides enough room for a plasma
sheath to naturally form around the probe. Here, the sheath is generated by the
interaction of the plasma within the boundary material.
In PROBEPIC it is assumed that the plasma is axisymmetric which allows the
calculation for the fields only in the r and z direction on the half plane as shown in
Figure7. In PROBEPIC, the computational grid plane can be divided into two sections;
25
interior grid points and surface grid points. Here, the interior grid points lie within the
plasma, probe conductor and insulator, while surface grid points lie at the edges of the
computational domain and on the surface of the probe conductor and insulator [2].
Figure7 also contains the geometry of the probe (in meters) for the reference case used
in this effort.
Figure 6: PROBEPIC computational domain layout created [2]
26
Figure 7: Field boundary conditions and probe geometry in PROBEPIC
2.2
Improvements/Updates
During the conversion of PROBEPIC, several complications occurred. These are
described in this section, and their solutions are presented. It was thought that early
discrepancies between the C++ (and FORTRAN) version and the original C version
occurred because of the technology differences between when the code was written in
(in 1996) and the present or because the available electronic copy of the code was not
the lastest version of PROBEPIC published in Markusic’s thesis. With several corrections
to the available electronic version of the code, similar results to the original data
presented in Markusic’s thesis [2] were obtained. These corrections are detailed below.
27
One of the problems occurred with “grid.c”, which is the code used to generate the
computational grid. Two of the computational domain parameters, “cell1_count” and
“cell6_count”, were hardcoded in this program instead of being set as parameters in
“parameter.c”, which is the code used to specify necessary parameters used in other
sub-programs to establish the initial plasma conditions. Before the correction of
“parameter.c”, the value of “cell-width”, as originally set in “parameter.c”, leads to the
domain illustrated below in Figure8, where the grid did not fill the domain. However,
the solution was applied by setting the cell_width equal to length divided by
ngp_horize-1, covering the entire domain by the grid as shown in Figure9.
Figure 8: The computational grid in PROBEPIC (before correction)
28
Figure 9: The computational grid after the correction of grid.c in PROBEPIC
The next issue was found in “inject_part.c”, which is a sub-program used to
move (i.e. flux) the thermal particles across boundaries into the computational domain.
The variable “count” had been constantly incremented and used as index in order to
browse the array of uniform random numbers. At a certain point, “count” exceeds the
dimension of “random_num” regardless of how big this was set. The solution to this
problem was to constantly check the value of “count” and reset it to zero when it
reaches the dimension of “random_num”, the length of the random number array.
Another issue occurred in “charge_weight.c”, which is used to weight charges on
particles to the grid points. During the process of a call to “locatex” and “locatey” (used
to determine which cell a particle is in), erroneous results for points near the upper
29
bound of the grid were obtained. The solution is achieved by decreasing the value of
second input parameter by one as locatex(.,ngp2-1,.,.) and locatey(.,ngp1-1,.,.), properly
defining the boundary.
After these corrections were implemented, the code started running and the
individual sub-programs gave the anticipated results. Note that some sub-programs
were run separately in test programs to observe if they work properly. For example, did
the particle velocity distribution function used in the injector provide the correct
distribution.
However, when the subroutines were combined together under a main-program
and run to create the simulation, the result was not as it was expected to be. Although
the results did not match those in Markusic’s thesis, they followed an astonishing path;
each point on the voltage-current characteristic was a factor of approximately -5/3
from the expected value (Markusic’s result). At first the scale difference was believed to
be the result of an erroneous data entry, e.g. a wrong geometrical dimension such as the
probe length or width, the computational domain, etc., or/and maybe a wrong physical
constant. To address this issue, the values of ngp1, ngp2, numpart, NPTS and
V_TABLE_RES, etc. in “lprobe60.h”, which is one of the header files in PROBEPIC
30
regarding probe geometry were parametrically changed to observe the level of
difference occurring in the program. Using this insight, some parameters were adjusted
and the difference between the new simulation results and Markusic’s results became a
factor of -1. The change from -5/3 to -1 is attributed to corrections in erroneous
geometry input. It was tempting to assume that the -1 occurred because of a matter of
sign convention in voltage-current characteristic, but the author was unsatisfied with
this explanation.
At this point the original PROBEPIC code was reexamined, and some more issues
were found in the logic of the program. The first of which was that the codes was not
correctly detecting when the particle lies inside or outside the computational domain.
This was resolved by making improvements in grid generation allowing higher density
of cells in the vicinity of the probe.
However, in the original program, an important approximation had not been
taken into consideration. A particle was considered to hit the probe when it falls exactly
in a cell on the probe. In this case, probe current is generated, otherwise it is not.
However, the high speed electrons and ions that have trajectories intersecting the probe
and not residing in on a probe cell at the end of a time step were not being included in
31
the probe current. This was resolved by adding a model to extend the particle trajectory
in a straight line (i.e. particle ray tracing) to determine if it intersects the cylinder
represented by the probe during a time step, see Figure10. With the inclusion of the
particles intersecting the probe, the code started giving the correct results without an
arbitrary -1 factor.
Figure 10: Particle ray tracing shown in animation window
32
While working on this particle collection issue, another logic issue was
identified involving the particle injection. As mentioned previously, the PROBEPIC
computational domain is a cylinder. The particle injection routine, “inject_part()”, was
intended to place new particles on the domain boundaries from six directions, i.e.
top/bottom, front/back, left side/right side, by using a virtual rectangular domain
containing the cylinder; the particles were placed on the faces of this rectangular
domain as shown in Figure11.
Figure 11: The rectangular domain in the original PROBEPIC
33
The particles will next travel for a “dt” interval of time, and then their new
positions are checked to be inside/outside cylindrical domain. For a low speed particle,
the “dt” interval may be too short for the particle to reach the cylindrical domain. For
this case the particle is not actually injected into the domain, despite the initial
intention. Only the particles fast enough to reach the domain will be injected into the
cylinder, i.e. low velocities in the velocity distribution of the injected particles were
underrepresented. Therefore, the “inject_part()” routine was modified to inject the new
particles directly along the cylindrical domain boundary over an angle of 90 degrees as
illustrated in Figure12.
Figure 12: Cylindrical domain boundary with particle injection over 90 degrees
34
Then the rest of the circle will be covered by particles injected from three remaining
sides of the virtual rectangular domain.
Additionally, a small issue was identified with the particle movement routine. In
the original code, the particle movement under constant acceleration was implemented
in the “mover()” routine using the equation below
S(t) = S0 + V0 t + A0 t 2
⃗ )/m in general and (e ∗ ⃗E)/m for the present case with no
where A0 is (e ∗ ⃗E + υxB
applied B - field. This equation was replaced with the following equation
t2
S(t) = S0 + V0 t + A0
2
The time step parameter, “dt”, was chosen empirically to satisfy the condition that a
particle not travel a distance greater than a cell size. This requirement is fulfilled on
average.
Another item that was adjusted was that almost all of the parameters that
control the simulation process that were hard coded in the original routine
“parameter()” were moved to on an editable input file, “parameter.txt”, giving the user
the option of performing multiple simulations without the requirement of recompiling
the program each time it is run. A sample “parameter.txt” input file is listed in Table3.
35
One can easily see that beside each numerical value contained in “parameter.txt”,
explanation of the input is provided.
Table 3: A sample of editable parameter.txt
0.1
percent of particles displayed in animation, 0.0,...,100.0 ; 0.0 means
NO animation
0.00003125 [m] probe radius
45.92
probe ASPECT RATIO: length/radius
3
num. of cells per probe radius
40000
initial num. of ions in domain
6
num. of injected ions from BACK at each time step
6
num. of injected ions from FRONT at each time step
12
num. of injected ions from SIDE at each time step
40000
initial num. of electrons in domain
5
num. of injected electrons from BACK at each time step
5
num. of injected electrons from FRONT at each time step
8
num. of injected electrons from SIDE at each time step
2.0
T[eV] temperature
1.0e9
[cm^(-3)] density number
1.673e-27
m_ion
1.6e-19
q_ion
9.11e-31
m_elec
-1.6e-19
q_elec
0.019
[sec] wpdt
0.0
plasma potential
36
Additionally, the original program did not allow the probe to be divided into
more than one cell in the radial direction. The routine “grid()” was modified to allow
this to be increased, allowing the grid to be denser in the vicinity of the probe than the
rest of the domain. A screen shot from the animation window is given in Figure13,
which shows the electrons/ions trajectories, the locations where the particles are
injected, and where they hit the probe.
The new version of the PROBEPIC now allows the user to update the code to run
new geometries and plasma conditions without modifying and recompiling the code;
the author put almost all parameters that control the simulation into the input file,
“parameter.txt”. The user can simply change the parameter(s), such as “wpdt” for time
step, or vertical grid density with “number of cells per probe radius”, etc. in the input
file, and then the user can launch the PROBEPIC.
Note that, the horizontal (axial) grid density is hardcoded to 72 cells (73 points)
in the file “lprobe60.cpp (lprobe60.F90)” by the variable called “ngp2”. If the horizontal
grid density change is desired, then “ngp2” must be changed to a new value, and the
program must then be re-compiled. This is the only situation when a re-compile is
needed, and it can be done easily. If a change in the vertical grid is desired, then the 4th
37
numerical value in the text file, “parameter.txt”, should be changed from 1 to 2, or 3, or
4, or more.
Figure 13: The locations where the particles injected and hit the probe
38
2.3
Comparison of New/Old Langmuir Simulation
The first version of the PIC code (PROBEPIC) has been updated and converted
into C++ and FORTRAN. Both the C++ and FORTRAN versions of PROBEPIC are running
properly together with their subroutines to provide results that are close to the
theoretical model of the probe presented in Ref.2. The simulation results of the original
and updated versions of PROBEPIC are similar to each other, providing verification that
the code conversion was achieved properly. A comparison of the voltage-current
characteristics is shown in Chapter III.
39
CHAPTER III
Results
In the original version of the PROBEPIC code, the simulations were conducted in
a quiescent plasma in the OLM limit. Similarly, in the updated version of the code, the
plasma conditions were also chosen to correspond to the orbital limited motion (OLM)
domain used in Markusic’s thesis [2]. Theoretical results are readily available for this
case. By comparing the new results with the results from the original version and the
theoretical results for an ideal probe, the effectiveness of the updated PROBEPIC can be
evaluated.
In Markusic’s thesis [2], simulations were conducted both for hydrogen plasma
(ion to electron mass ratio of 1836) and a light ion simulation with an artificial ion to
electron mass ratio of 100. The “Light ion” experiment was conducted to verify that the
ion collection was being properly modeled with fewer time steps. However, an objective
of this thesis was to determine if the code is error free and the conversion to FORTRAN
was successful. For the updated version, due to time constraints only the Hydrogen
plasma (1836 mass ratio) experiment was conducted to verify that the C++ and
40
FORTRAN versions of the code run properly side by side. Figure14 illustrates the
updated simulation results on a voltage-current characteristic plot. The dashed and
solid curves are the theoretical results for an ideal Langmuir probe in the OLM limit.
Simulation results reported by Markusic [2] are shown as open squares and circles. The
results from the updated code are shown as solid red squares.
As can be seen in Figure14, if the probe potential is less than the plasma
potential, then the PROBEPIC results show good agreement with the Langmuir theory.
Figure 14: Voltage-Current (V-I) characteristics of original and updated PROBEPIC code
41
On the other hand, the PROBEPIC results diverge from the expected theoretical results
in the electron saturation region (voltages above the plasma potential). The reason for
this is that the Langmuir theory always gives the maximum current that could be
measured since the ideal probe is assumed. Therefore, the PROBEPIC data should always
appear below the upper orbital limited motion (OLM) threshold [2].
According to conclusion drawn by Markusic [2], the disagreement between the
Langmuir theory and PROBEPIC had a physical reason. The Langmuir theory was
developed for an infinitely long cylindrical probe. However, for PROBEPIC and the
laboratory experiments [3], the finite length probes were used. This suggests that the
probe aspect ratio may produce a significant difference in the voltage-current
characteristic. Also the sharp edges on finite length probe tips on conductors may
produce an intense electric field in the vicinity of the tip that more efficiently draws in
charged particles than realized in an “infinite” length probe. Note that since both probe
types exhibit orbital limited motion (OML) current collection, the resulting voltagecurrent characteristic does not have a sharp knee when the probe potential is equal to
the plasma potential. Rather, there occurred a very smooth curve from positive to
negative, as is expected (see Chapter I).
42
However, it should be noted that it appears that the enhancements made to
PROBEPIC in the current effort did produce an improvement in the accuracy of the
predictions. By comparing the new data points to the earlier data points the gap
between the predictions and the Langmuir theory is reduced by about 30% in the
electron saturation region. While some of this improvement appears to be from the
enhanced particle collection and injection models, the difference between PROBEPIC
predictions and Langmuir theory appears to be sensitive to grid density.
The lower grid density was examined to determine if the same results could be
obtained in a shorter computational time. Grid density was lowered 50 percent as seen
in Figure15. When the grid density was reduced, it was observed that the predicted
current changed significantly, diverging away from the theoretical points. Although the
grid change lessened the computational time, this experiment showed that lowering the
grid density had a big impact on the result. Because of this observed sensitivity, it is not
known if the baseline density is sufficient to provide the best current prediction. To
determine the optimum grid density, it will be necessary to perform additional work.
In this future work, the number of injected particles should also be examined to
determine the impact of this parameter on the current prediction, since the low grid
43
density experiment also showed that the running time depends on the number of
particles in the domain, rather than just the number of mesh points. This sensitivity is
the result of the need to move and test for position of all the particles at each time step.
Figure 15: Probe grid with 50 percent lower density
44
CHAPTER IV
Conclusions and Recommendations
4.1
Conclusions
Markusic created the Particle-in-Cell code (PROBEPIC) to simulate the behavior
of finite length Langmuir probes via particle simulation. In the present thesis work,
Markusic’s code has been converted from C into C++ and FORTRAN. During conversion
to C++ and FORTRAN, several errors in the available electronic version of Markusic’s
code were identified and corrected. After the code conversions, the simulation of
hydrogen plasma with parameters that correspond to the OML case showed that new
version of PROBEPIC agrees well with Markusic’s original results. The correction of
errors and code improvements in the updated version resulted in simulated currents
closer to theory in the electron saturation region of the V-I characteristic, a region with
the largest departure from theory. Also, the updated version is more user friendly with
the addition of a user input file. Some changes were also made to allow the code to be
more applicable to other probe geometries.
45
4.2
Recommendations
The updated version of PROBEPIC can be further improved in future work by
enhancing code speed with the algorithm refinements. In addition modifying the code
to include a hybrid model, i.e. implementing a fluid model for the electrons eliminating
the need for two different time scales [2] should be considered. Additional simulations,
such as simulating plasma/probe interactions in the thin sheath limit, are also of
interest. Conducting simulations with alternate Langmuir probe geometries such as
increasing the probe aspect ratio, and applying the code to other probe types, such as a
Faraday probe, are recommended.
Finally, the lower grid density (50 percent less) was examined to see if the same
results could be obtained in a shorter computational time. The results showed that
lowering the grid density has a significant effect on the current prediction, causing it to
diverge away from the theoretical results. In the future, different grid densities can be
examined to find the optimum grid density to provide the best match between the
predicted data points and the theoretical values. For this kind of experiments, higher
capacity computer systems will be required.
46
BIBLIOGRAPHY
47
[1]H.Mott-Smith, Jr and I. Langmuir. Phys.Rev.28, 27. 1926.
[2]Markusic, Thomas. Particle Simulation of a Langmuir Probe in Quescient and
Flowing Plasmas. Masters Thesis. Tullahoma, 1996.
[3]Keefer, D. and D.D. Semak. "Measurements of Radial and Axial Distributions of Ion
Thruster Plasma Parameters Using a Langmuir Probe (unpublished).
[4]Merlino, Robert L. "Understanding Langmuir Probe Current-Voltage Characteristics."
2007.
[5]Hoag, J.B. and S.A. Korff. Electron and Nuclear Physics. New York: Van Norstrand,
1952.
[6]Fei, Yee Yue. "The I-V Ramp Circuit for Langmuir Probe." Techpos . 2004.
[7]Birdsall, C.K. and A.B. Langdon. Plasma Physics via Computer Simulation. New
York: Adam Hilger, 1991.
[8]Azooz, A. A. Four free parameter emprical parametrization of glow discharge
Langmuir peobe data. Review of Scientific Instruments. American Institute of
Physics, October 3, 2008.
[9]Wang, J. and J. Brophy. 3-D Monte-Carlo Particle-in-Cell Simulations of Ion
Thruster Plasma Interactions. San Diego, CA: AIAA, 1995.
48
APPENDICES
49
Appendix-I
<<<The PROBEPIC code will be added in this section>>>
50
Vita
Denise Su (Densu) A. Aktas was born in 1987 in Istanbul. She finished her high
school in 2001 and entered the university to study Electronic Engineering degree at
Marmara University in Istanbul. She graduated from the university in 2005. During her
studying engineering, she won several scholarships from honor organizations both in
her country and abroad. Then she got her second degree in Electromechanical
Engineering Technology at MTSU in 2007 to combine her electrical and mechanical
knowledge for technical applications. Because of her professors’ recommendations, she
entered the graduate school at UT Space Institute in 2007 to pursue Masters Degree.
She completed the degree requirements for both Aviation Systems and Mechanical
Engineering. She plans to pursue her PhD degree also in Mechanical Engineering
specializing in electric propulsion.
51